Bound-based approximations in multistage stochastic programming: Nonanticipativity aggregation

This paper considers bounding techniques within multistage stochastic programming and their computational issues. First, a brief overview of bound-based approximations is provided within the context of two-stage stochastic programs. Then, a new methodology for approximating multistage stochastic programs is developed using a process called non-anticipativity aggregation. The method advocates relaxing nonanticipativity requirements of the decisions up to a certain second-order aggregation to compute lower bounds. Then, this relaxation is refined progressively towards solving the stochastic program. Preliminary computational results are presented.